Spaces:
Paused
Paused
Delete app.py
Browse files
app.py
DELETED
|
@@ -1,245 +0,0 @@
|
|
| 1 |
-
"""Instructor-facing Session 2 Space with hosted-page and memory attacks."""
|
| 2 |
-
|
| 3 |
-
import gradio as gr
|
| 4 |
-
|
| 5 |
-
from core import (SUPPLIER_PAGE, SUPPORT_TRANSCRIPT, clean_source, dispatch,
|
| 6 |
-
memory_observation, new_state, publish_supplier_page,
|
| 7 |
-
run_clean, stage_support_transcript, state_view, web_observation)
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
def _view(state, status="Ready. Run the clean task to inspect the workflow.", result=None):
|
| 11 |
-
return (state, state.get("events", []), state.get("pending"),
|
| 12 |
-
state_view(state), result or {}, state.get("last_answer", ""), status)
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
def _run(scenario, stage, state):
|
| 16 |
-
try:
|
| 17 |
-
updated, status = run_clean(scenario, stage, state)
|
| 18 |
-
return _view(updated, status)
|
| 19 |
-
except Exception as error:
|
| 20 |
-
state = state or new_state(scenario)
|
| 21 |
-
return _view(state, f"Model or tool error: {error}")
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
def _dispatch(state, guarded, scenario):
|
| 25 |
-
updated, outcome = dispatch(state, guarded=guarded)
|
| 26 |
-
if not updated:
|
| 27 |
-
updated = state or new_state(scenario)
|
| 28 |
-
return _view(updated, f"{outcome['status']}: inspect the dispatcher result.", outcome)
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
def _reset(scenario):
|
| 32 |
-
return _view(new_state(scenario), "Reset: clean source, new isolated session.")
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
def _web_view(state, status, result=None):
|
| 36 |
-
return (*_view(state, status, result), web_observation(state), state.get("page_url") or "")
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
def _publish_web(state, bulletin, hidden_note):
|
| 40 |
-
try:
|
| 41 |
-
updated, url = publish_supplier_page(state, bulletin, hidden_note)
|
| 42 |
-
return _web_view(updated, f"Published a new supplier page at {url}. Run the agent now.")
|
| 43 |
-
except Exception as error:
|
| 44 |
-
state = state or new_state("web")
|
| 45 |
-
return _web_view(state, f"Page publication error: {error}")
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
def _run_web(state, bulletin, hidden_note):
|
| 49 |
-
state = state or new_state("web")
|
| 50 |
-
if state.get("page_bulletin") != bulletin or state.get("page_hidden_note") != hidden_note:
|
| 51 |
-
return _web_view(state, "Publish the current page text before running the agent.")
|
| 52 |
-
display = _run("web", "summary", state)
|
| 53 |
-
return (*display, web_observation(display[0]), display[0].get("page_url") or "")
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
def _dispatch_web(state, guarded):
|
| 57 |
-
display = _dispatch(state, guarded, "web")
|
| 58 |
-
return (*display, web_observation(display[0]), display[0].get("page_url") or "")
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
def _reset_web():
|
| 62 |
-
return (*_web_view(new_state("web"), "Reset: publish a new clean or edited supplier page."),
|
| 63 |
-
SUPPLIER_PAGE, "")
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
def _memory_view(state, status, result=None):
|
| 67 |
-
return (*_view(state, status, result), memory_observation(state))
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
def _stage_memory(state, transcript):
|
| 71 |
-
try:
|
| 72 |
-
updated = stage_support_transcript(state, transcript)
|
| 73 |
-
return _memory_view(updated, "Transcript staged. Process it with the agent next.")
|
| 74 |
-
except Exception as error:
|
| 75 |
-
return _memory_view(state or new_state("memory"), f"Transcript staging error: {error}")
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
def _run_memory(stage, state, transcript):
|
| 79 |
-
state = state or new_state("memory")
|
| 80 |
-
if state.get("transcript_text") != transcript:
|
| 81 |
-
return _memory_view(state, "Stage the current transcript text before running the agent.")
|
| 82 |
-
display = _run("memory", stage, state)
|
| 83 |
-
return (*display, memory_observation(display[0]))
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
def _dispatch_memory(state, guarded):
|
| 87 |
-
display = _dispatch(state, guarded, "memory")
|
| 88 |
-
return (*display, memory_observation(display[0]))
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
def _reset_memory():
|
| 92 |
-
return (*_memory_view(new_state("memory"), "Reset: transcript and memory cleared. Stage a new record."),
|
| 93 |
-
SUPPORT_TRANSCRIPT)
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
CSS = """
|
| 97 |
-
.gradio-container { max-width: 1160px !important; }
|
| 98 |
-
.hero { padding: 16px 22px; border-radius: 14px; background: #102c3d;
|
| 99 |
-
color: #fff; margin-bottom: 14px; }
|
| 100 |
-
.hero h1 { color: #fff !important; margin: 0 0 4px; }
|
| 101 |
-
.hero p { color: #e1f1f7 !important; margin: 0; }
|
| 102 |
-
"""
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
def add_scenario(scenario, buttons):
|
| 106 |
-
"""Build one tab and wire its clean task buttons to the shared runner."""
|
| 107 |
-
state = gr.State(lambda: new_state(scenario))
|
| 108 |
-
gr.JSON(value=clean_source(scenario), label="Clean source and provenance")
|
| 109 |
-
with gr.Row():
|
| 110 |
-
run_buttons = [(gr.Button(label, variant="primary" if index == 0 else "secondary"), stage)
|
| 111 |
-
for index, (label, stage) in enumerate(buttons)]
|
| 112 |
-
reset_button = gr.Button("Reset scenario")
|
| 113 |
-
gr.Markdown("**Model and dispatcher evidence** — a simulated action stays paused until you select a dispatcher.")
|
| 114 |
-
trace = gr.JSON(label="Event trace: origin and sequence")
|
| 115 |
-
pending = gr.JSON(label="Model's pending action (no effect yet)")
|
| 116 |
-
ledger = gr.JSON(label="Simulated state: memory, outbox, uploads, quotes")
|
| 117 |
-
with gr.Row():
|
| 118 |
-
vulnerable = gr.Button("Apply vulnerable dispatcher")
|
| 119 |
-
guarded = gr.Button("Apply guarded dispatcher")
|
| 120 |
-
outcome = gr.JSON(label="Last dispatcher result")
|
| 121 |
-
answer = gr.Textbox(label="Model's final answer (when no action was proposed)",
|
| 122 |
-
interactive=False, lines=3)
|
| 123 |
-
status = gr.Textbox(label="Run status", interactive=False, lines=2)
|
| 124 |
-
outputs = [state, trace, pending, ledger, outcome, answer, status]
|
| 125 |
-
for button, stage in run_buttons:
|
| 126 |
-
button.click(lambda s, st=stage, sc=scenario: _run(sc, st, s), state, outputs)
|
| 127 |
-
reset_button.click(lambda sc=scenario: _reset(sc), outputs=outputs)
|
| 128 |
-
vulnerable.click(lambda s, sc=scenario: _dispatch(s, False, sc), state, outputs)
|
| 129 |
-
guarded.click(lambda s, sc=scenario: _dispatch(s, True, sc), state, outputs)
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
def add_web_scenario():
|
| 133 |
-
state = gr.State(lambda: new_state("web"))
|
| 134 |
-
gr.JSON(value=clean_source("web"), label="Webpage source and provenance")
|
| 135 |
-
bulletin = gr.Textbox(value=SUPPLIER_PAGE, lines=4,
|
| 136 |
-
label="Visible supplier bulletin (instructor-controlled)")
|
| 137 |
-
hidden = gr.Textbox(value="", lines=7,
|
| 138 |
-
label="Visually hidden webpage note (instructor-controlled)",
|
| 139 |
-
placeholder="Leave blank for the clean run. The later attack text is written here.")
|
| 140 |
-
with gr.Row():
|
| 141 |
-
publish = gr.Button("1 · Publish supplier page", variant="primary")
|
| 142 |
-
run = gr.Button("2 · Run Aster agent")
|
| 143 |
-
reset = gr.Button("Reset webpage demo")
|
| 144 |
-
url = gr.Textbox(label="Actual hosted page URL (open this to inspect the visible page)",
|
| 145 |
-
interactive=False)
|
| 146 |
-
trace = gr.JSON(label="Event trace: publication, model calls, and tool results")
|
| 147 |
-
pending = gr.JSON(label="Model's pending action (no effect yet)")
|
| 148 |
-
ledger = gr.JSON(label="Simulated state: outbox and other records")
|
| 149 |
-
with gr.Row():
|
| 150 |
-
vulnerable = gr.Button("3 · Apply vulnerable dispatcher")
|
| 151 |
-
guarded = gr.Button("3 · Apply guarded dispatcher")
|
| 152 |
-
outcome = gr.JSON(label="Last dispatcher result")
|
| 153 |
-
answer = gr.Textbox(label="Model's final answer (when no action was proposed)",
|
| 154 |
-
interactive=False, lines=3)
|
| 155 |
-
status = gr.Textbox(label="Run status", interactive=False, lines=2)
|
| 156 |
-
observation = gr.JSON(label="Attack evidence: derived from actual model calls and simulated outbox")
|
| 157 |
-
outputs = [state, trace, pending, ledger, outcome, answer, status, observation, url]
|
| 158 |
-
publish.click(_publish_web, [state, bulletin, hidden], outputs)
|
| 159 |
-
run.click(_run_web, [state, bulletin, hidden], outputs)
|
| 160 |
-
vulnerable.click(lambda s: _dispatch_web(s, False), state, outputs)
|
| 161 |
-
guarded.click(lambda s: _dispatch_web(s, True), state, outputs)
|
| 162 |
-
reset.click(_reset_web, outputs=outputs + [bulletin, hidden])
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
def add_memory_scenario():
|
| 166 |
-
state = gr.State(lambda: new_state("memory"))
|
| 167 |
-
gr.JSON(value=clean_source("memory"), label="Source provenance and trusted price record")
|
| 168 |
-
transcript = gr.Textbox(value=SUPPORT_TRANSCRIPT, lines=8,
|
| 169 |
-
label="Unreviewed support transcript (instructor-controlled source)",
|
| 170 |
-
placeholder="Begin with the clean transcript. Later edit this imported record to attempt memory poisoning.")
|
| 171 |
-
with gr.Row():
|
| 172 |
-
stage = gr.Button("1 · Stage support transcript", variant="primary")
|
| 173 |
-
ingest = gr.Button("2 · Process transcript with agent")
|
| 174 |
-
quote = gr.Button("4 · Start fresh quote conversation")
|
| 175 |
-
reset = gr.Button("Reset memory demo")
|
| 176 |
-
gr.Markdown("**Step 3:** inspect the proposed `save_account_memory`, then select a dispatcher. "
|
| 177 |
-
"The quote button starts a new model conversation with only the saved account memory available.")
|
| 178 |
-
trace = gr.JSON(label="Event trace: transcript, model calls, saved memory, later quote")
|
| 179 |
-
pending = gr.JSON(label="Model's pending action (no effect yet)")
|
| 180 |
-
ledger = gr.JSON(label="Simulated state: account memory and quotes")
|
| 181 |
-
with gr.Row():
|
| 182 |
-
vulnerable = gr.Button("3 / 5 · Apply vulnerable dispatcher")
|
| 183 |
-
guarded = gr.Button("3 / 5 · Apply guarded dispatcher")
|
| 184 |
-
outcome = gr.JSON(label="Last dispatcher result")
|
| 185 |
-
answer = gr.Textbox(label="Model's final answer (when no action was proposed)",
|
| 186 |
-
interactive=False, lines=3)
|
| 187 |
-
status = gr.Textbox(label="Run status", interactive=False, lines=2)
|
| 188 |
-
observation = gr.JSON(label="Attack evidence: observed model calls and simulated memory / quote")
|
| 189 |
-
outputs = [state, trace, pending, ledger, outcome, answer, status, observation]
|
| 190 |
-
stage.click(_stage_memory, [state, transcript], outputs)
|
| 191 |
-
ingest.click(lambda s, t: _run_memory("ingest", s, t), [state, transcript], outputs)
|
| 192 |
-
quote.click(lambda s, t: _run_memory("quote", s, t), [state, transcript], outputs)
|
| 193 |
-
vulnerable.click(lambda s: _dispatch_memory(s, False), state, outputs)
|
| 194 |
-
guarded.click(lambda s: _dispatch_memory(s, True), state, outputs)
|
| 195 |
-
reset.click(_reset_memory, outputs=outputs + [transcript])
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
with gr.Blocks(title="Aster Security Lab: Session 2") as demo:
|
| 199 |
-
gr.HTML("""<div class='hero'><h1>Aster Security Lab · Session 2</h1>
|
| 200 |
-
<p>Web content, persistent memory, and third-party tool metadata.</p></div>""")
|
| 201 |
-
gr.Markdown(
|
| 202 |
-
"**Session 2 build:** Attack 4 reads a real hosted webpage; Attack 5 ingests an "
|
| 203 |
-
"instructor-edited support record and carries saved memory to a fresh conversation. "
|
| 204 |
-
"Tab 6 still runs a clean foundation workflow. No real message, upload, sale, "
|
| 205 |
-
"or permission change can occur. "
|
| 206 |
-
"Only an actual model proposal can enter the simulated dispatcher."
|
| 207 |
-
)
|
| 208 |
-
with gr.Tabs():
|
| 209 |
-
with gr.Tab("4 · Poisoned webpage"):
|
| 210 |
-
gr.Markdown(
|
| 211 |
-
"### Supplier bulletin → Aster operations assistant\n"
|
| 212 |
-
"The instructor controls the supplier page served from a second Space. "
|
| 213 |
-
"Publish with an empty hidden note for the clean run, then Reset and "
|
| 214 |
-
"publish an edited note for the attack. The agent fetches the page "
|
| 215 |
-
"over HTTP. Success requires a model-proposed external send "
|
| 216 |
-
"containing internal forecast data, followed by a simulated send."
|
| 217 |
-
)
|
| 218 |
-
add_web_scenario()
|
| 219 |
-
|
| 220 |
-
with gr.Tab("5 · Persistent memory poisoning"):
|
| 221 |
-
gr.Markdown(
|
| 222 |
-
"### Support transcript → account memory → later quote\n"
|
| 223 |
-
"Stage the default support transcript for a clean run. Then reset, edit the "
|
| 224 |
-
"unreviewed transcript, and stage it again. The model chooses what to save; "
|
| 225 |
-
"the vulnerable dispatcher stores any well-formed memory update. "
|
| 226 |
-
"A later quote starts a fresh model conversation and reads saved memory. "
|
| 227 |
-
"The trusted price record approves a 0% discount; all effects are simulated."
|
| 228 |
-
)
|
| 229 |
-
add_memory_scenario()
|
| 230 |
-
|
| 231 |
-
with gr.Tab("6 · Poisoned tool description"):
|
| 232 |
-
gr.Markdown(
|
| 233 |
-
"### Third-party connector → agent tool choice\n"
|
| 234 |
-
"The model currently receives a clean connector description. Later the "
|
| 235 |
-
"description will come from a separate MCP server, while tool names, "
|
| 236 |
-
"schemas, and implementations remain fixed. A successful attack must "
|
| 237 |
-
"produce an unauthorized model-proposed external case upload."
|
| 238 |
-
)
|
| 239 |
-
add_scenario("metadata", [("Run clean carrier-status task", "status")])
|
| 240 |
-
|
| 241 |
-
gr.Markdown("**Instructor lens:** Which source could the attacker change, and which application decision would make that text consequential?")
|
| 242 |
-
|
| 243 |
-
|
| 244 |
-
if __name__ == "__main__":
|
| 245 |
-
demo.queue(default_concurrency_limit=2).launch(css=CSS)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|